More realistically, we're going to type something like "pa[Tab] ball Jo[Tab]", pressing the Tab key whenever Ubiquity suggests to us the word/thing that we want. After the second Tab, it will autocomplete the entire grammatically-correct sentence ("Pass the ball to Jono.") for us to verify before we confirm by pressing Enter.

I think this gives some flexibility to you in what types of things you suggest.

It's true that, just like in real life with other people, the utterances we make are very often ambiguous. However, just like we ask each other for confirmations when there are ambiguities, Ubiquity can do the same. For that, I think Aza's solving the it problem post is a really interesting feedback model and I like it a lot. Building a statistical "user grammar" on top of the base grammar is an interesting suggestion that we should look at.

Recommending new verbs from the herd or elsewhere is something I want to think about in the future as well. I look forward to your comments in the future! ^^

So learnability and usability should be a two-way approach - self-modifying - between user and Ubiquity: the user experience improves if the Ubiquity learns what the user wants AND the syntactic expression used, building a user "grammar" on a default "grammar".

Discoverability paradigm could be the other face of learnability from the user point of view: what is the command I should use for? Try and error approach: the user types a verb and Ubiquity could be suggest a subscribed verb or a new verb at all (listed in an online catalog - herd).

I look forward to reading your next interesting posts.